Surely over time since Einstein started on BOINC the trend has been inflationary, or else they wouldn't be trying to reduce credits now.
Can you prove that?
If you can definitively prove it, I'm all eyes...
What all of you in favor of doing this gerrymandering (that's all it really is) seem to like to do is look at "on average". It's a great concept, and in certain aspects it is the right view to take. You (general sense) appear to be either unwilling or unable to change the focus to what happens inside of a single project when there is a credit shift.
These are all going to be hypothetical numbers. Just try to follow the concept. I tried this once with the widget example. You don't appear to have understood. Either that or you didn't see it or you dismissed it out of hand.
Host A - runs Project A from June 1999 to February 2001. Tasks are awarded 1 unit each. Tasks took 1 hour a piece. Host A processed a total of 1000 tasks. Host A is awarded 1000 units of credit for 1000 hours of processing time. Host A has a Credit per CPU Hour of 1. Host A returns to processing tasks in 2008. Tasks are now awarded 0.85 units each. Host A processes an additional 1000 tasks, but tasks now take 1.5 hours each due to increased science level in the application.
Host A totals: 1820 credits, CPCH of 0.78333 across 2000 tasks
Host B - processes tasks during 2008. Is awarded 0.85 per task. Processes 2000 tasks. Each task takes 1.5 hours.
Host B totals: 1700 credits, CPCH of 0.56667 across 2000 tasks.
*Host A is both higher in credits and in CPCH.
*Host B technically "did more science", since all 2000 tasks were with the application that performed more science per task.
*Each host processed exactly 2000 tasks
*Host A took 2500 hours to process 2000 tasks, while Host B took 3000 hours.
Based on credits and credits alone, Host A would have a higher ranking. This is exactly how the BOINCstats logo that will be at the bottom of this post works.
Now, looking at this from a science perspective, Host B should have the higher ranking. Through no fault of their own, their owner participated in the project during a span of time with average lower credit per task than Host A's total span of time had.
Finally, if looking at this from the CPCH perspective, we're back to Host A as the host that should have the higher ranking, however it is skewed by the "unfair" advantage that it had over Host B due to the "inflated" credits from its' earlier participation.
This is what I was saying earlier, that the only way to guarantee that the by-credit rankings are accurate is if the two hosts in question ran during the same general credit scheme across the total time of their participation.
This is the problem with "The Cobblestone Approach" to have "equal" credits across projects. An unintended side-effect is that it can distort the rankings, making it "unequal" / "unfair" in its' own way...
Extending on Alinator's "cpcs = 'power'" example, this is exactly correct.
If an application improvement makes it to where a system processes a task faster, how exactly is that any different from:
*Overclocking
*Buying a faster cpu
*Adding more memory (if memory causes an increase in performance)
These things that a user could do to a host are not "punished" for their efficiency. Why are application improvements "punished"? Why is my overclocking of my CPU not "punished"?
So long as an honest effort is made to extend improvements to all platforms to which the improvements have technical capability, then the application improvement is "fairly distributed". The improvements here have not yet been "fairly distributed" between Linux and Windows. The improvements here may not apply to certain classes of processors. A Pentium 233 MMX cannot use SSE instructions - it doesn't know how. Is this a "punishment" to not extend the improvements to this class of processor? No, because the processor is not technically capable of utilizing the improvements. "Due diligence" / "honest effort" was still made by the project.
In some ways, CPP folks seem to indicate that they want the actual performance gap between their slower systems and today's faster systems to be reduced to some degree by a "fair" credit system.
Okay, I've been mostly ignoring this because you all are really bumming me out by turning my "happy I got a new task" thread into another credit argument. :-(
However, I can say that if you want to for an overall trend in either inflation or deflation of credits per unit of processor (work/time/whatever) then Climateprediction.net would make a pretty good "standard candle" for the purpose.
I'm not positive on this, but I believe they've not adjusted the credits there for around 2 years or so. (I seem to remember not too long after they introduced the first "coupled models" they figured out they'd mis-estimated the amount of processing they would take and they nudged the figure up some.) I went back and checked and I've been "paid" the same for all the work I've done for them since roughly January of 2007. I can also tell you that for almost 2 years I processed only Einstein (45%), CPDN (45%) and SETI (10% as a backup) and at that time, I seem to remember getting virtually identical RAC's for both E@H and CPDN. Certainly they were within a negligible difference of each other.
I'm not going to say that what CPDN pays is (right/wrong/high/low/good/bad), just simply that if you want to compare credit "today" vs credit "yesteryear" then they'd make a good comparison.
Okay, I've been mostly ignoring this because you all are really bumming me out by turning my "happy I got a new task" thread into another credit argument. :-(
Okay, I've been mostly ignoring this because you all are really bumming me out by turning my "happy I got a new task" thread into another credit argument. :-(
Consider it dropped...
Just so you know Brian, I respect your viewpoints on credit. I don't necessarily agree with everything you say, but you at least put thought into your arguments and for the most part keep things as a civil debate. (Not to mention that those with the first name Brian are among the finest on the planet. Ahem... it's mine as well. ;) )
It just seemed like this thread was about to take a left turn into crazy and I had kind of hoped it would just remain as a place for folks to step out and shout to the world that they were happy a new run had started! (Okay I didn't put THAT much thought into it, but it was either create the post or walk out of my front door and shout it to the world and my neighbors probably already think I'm weird enough as it is...)
In any case, they have a thread dedicated to credit discussions here and I'll read it occasionally and even chime in now and again but for the most part, I think all matters of credit are mostly beyond logical arguments because everyone has too strong an opinion about how they think things 'should' be done at this point.
(Not to mention that those with the first name Brian are among the finest on the planet. Ahem... it's mine as well. ;) )
My brother's name is Brian - does that mean that at least I come from a good family? ;)
Quote:
but it was either create the post or walk out of my front door and shout it to the world and my neighbors probably already think I'm weird enough as it is...)
If I did that my kids would simply think 'Dad's being wierd again' !
Quote:
I think all matters of credit are mostly beyond logical arguments because everyone has too strong an opinion about how they think things 'should' be done at this point.
Now THAT is one of the best points I've read in a while!
RE: Surely over time since
)
Can you prove that?
If you can definitively prove it, I'm all eyes...
What all of you in favor of doing this gerrymandering (that's all it really is) seem to like to do is look at "on average". It's a great concept, and in certain aspects it is the right view to take. You (general sense) appear to be either unwilling or unable to change the focus to what happens inside of a single project when there is a credit shift.
These are all going to be hypothetical numbers. Just try to follow the concept. I tried this once with the widget example. You don't appear to have understood. Either that or you didn't see it or you dismissed it out of hand.
Host A - runs Project A from June 1999 to February 2001. Tasks are awarded 1 unit each. Tasks took 1 hour a piece. Host A processed a total of 1000 tasks. Host A is awarded 1000 units of credit for 1000 hours of processing time. Host A has a Credit per CPU Hour of 1. Host A returns to processing tasks in 2008. Tasks are now awarded 0.85 units each. Host A processes an additional 1000 tasks, but tasks now take 1.5 hours each due to increased science level in the application.
Host A totals: 1820 credits, CPCH of 0.78333 across 2000 tasks
Host B - processes tasks during 2008. Is awarded 0.85 per task. Processes 2000 tasks. Each task takes 1.5 hours.
Host B totals: 1700 credits, CPCH of 0.56667 across 2000 tasks.
*Host B technically "did more science", since all 2000 tasks were with the application that performed more science per task.
*Each host processed exactly 2000 tasks
*Host A took 2500 hours to process 2000 tasks, while Host B took 3000 hours.
Based on credits and credits alone, Host A would have a higher ranking. This is exactly how the BOINCstats logo that will be at the bottom of this post works.
Now, looking at this from a science perspective, Host B should have the higher ranking. Through no fault of their own, their owner participated in the project during a span of time with average lower credit per task than Host A's total span of time had.
Finally, if looking at this from the CPCH perspective, we're back to Host A as the host that should have the higher ranking, however it is skewed by the "unfair" advantage that it had over Host B due to the "inflated" credits from its' earlier participation.
This is what I was saying earlier, that the only way to guarantee that the by-credit rankings are accurate is if the two hosts in question ran during the same general credit scheme across the total time of their participation.
This is the problem with "The Cobblestone Approach" to have "equal" credits across projects. An unintended side-effect is that it can distort the rankings, making it "unequal" / "unfair" in its' own way...
Extending on Alinator's "cpcs
)
Extending on Alinator's "cpcs = 'power'" example, this is exactly correct.
If an application improvement makes it to where a system processes a task faster, how exactly is that any different from:
*Buying a faster cpu
*Adding more memory (if memory causes an increase in performance)
These things that a user could do to a host are not "punished" for their efficiency. Why are application improvements "punished"? Why is my overclocking of my CPU not "punished"?
So long as an honest effort is made to extend improvements to all platforms to which the improvements have technical capability, then the application improvement is "fairly distributed". The improvements here have not yet been "fairly distributed" between Linux and Windows. The improvements here may not apply to certain classes of processors. A Pentium 233 MMX cannot use SSE instructions - it doesn't know how. Is this a "punishment" to not extend the improvements to this class of processor? No, because the processor is not technically capable of utilizing the improvements. "Due diligence" / "honest effort" was still made by the project.
In some ways, CPP folks seem to indicate that they want the actual performance gap between their slower systems and today's faster systems to be reduced to some degree by a "fair" credit system.
Is that "fair"?
RE: Can you prove that? If
)
Okay, I've been mostly ignoring this because you all are really bumming me out by turning my "happy I got a new task" thread into another credit argument. :-(
However, I can say that if you want to for an overall trend in either inflation or deflation of credits per unit of processor (work/time/whatever) then Climateprediction.net would make a pretty good "standard candle" for the purpose.
I'm not positive on this, but I believe they've not adjusted the credits there for around 2 years or so. (I seem to remember not too long after they introduced the first "coupled models" they figured out they'd mis-estimated the amount of processing they would take and they nudged the figure up some.) I went back and checked and I've been "paid" the same for all the work I've done for them since roughly January of 2007. I can also tell you that for almost 2 years I processed only Einstein (45%), CPDN (45%) and SETI (10% as a backup) and at that time, I seem to remember getting virtually identical RAC's for both E@H and CPDN. Certainly they were within a negligible difference of each other.
I'm not going to say that what CPDN pays is (right/wrong/high/low/good/bad), just simply that if you want to compare credit "today" vs credit "yesteryear" then they'd make a good comparison.
RE: RE: Can you prove
)
Consider it dropped...
RE: RE: RE: Can you
)
Just so you know Brian, I respect your viewpoints on credit. I don't necessarily agree with everything you say, but you at least put thought into your arguments and for the most part keep things as a civil debate. (Not to mention that those with the first name Brian are among the finest on the planet. Ahem... it's mine as well. ;) )
It just seemed like this thread was about to take a left turn into crazy and I had kind of hoped it would just remain as a place for folks to step out and shout to the world that they were happy a new run had started! (Okay I didn't put THAT much thought into it, but it was either create the post or walk out of my front door and shout it to the world and my neighbors probably already think I'm weird enough as it is...)
In any case, they have a thread dedicated to credit discussions here and I'll read it occasionally and even chime in now and again but for the most part, I think all matters of credit are mostly beyond logical arguments because everyone has too strong an opinion about how they think things 'should' be done at this point.
RE: (Not to mention that
)
My brother's name is Brian - does that mean that at least I come from a good family? ;)
If I did that my kids would simply think 'Dad's being wierd again' !
Now THAT is one of the best points I've read in a while!
Regards
Rod (brother of another Brian)